Tim, Flashdrives are not going to make any surprising leaps and bounds due Channel performance. The concurrent throughput of Flashdrives is limited by the Back-end topology of each vendor's storage, and has very little to do with the Host Channel connections to cache. If a customer wants/needs to take advantage of faster channel speeds that then Permacache and FlashAccess would be a much better choice where the size of the data is small. This is true SSD performance as we knew it a decade and a half ago.
The box-wide performance of Read Cache miss on Flashdrives will not be much better than the Cache Read Miss performance of a fully popped storage controller. The advantage of Flashdrives is the incredible access density you can support so you don't have to install a truck load of short stroked disk drives for a cache unfriendly workload. All storage platforms available for MF now would need huge engineering changes to support the peak random throughput available from more than 16-32 flash drives. Just eight of these babies can do 60,000 cache miss IOPS. The disk controller needs to support 240,000 cache miss IOPS if you want to max out more than 32 Flashdrives. The beauty of Flashdrives is that is that you can roll up 1TB of cache unfriendly datasets into a single 7+1 array of Flashdrives. If that application had a 10,000 cache miss IOPS requirement you would need to spread this 1TB over 50-60 short stroked disk drives. Faster Channels won't change any of this. Ron > > Another thing to keep in mind is that Solid State Disk (SSD) is shipping > from all the major enterprise storage vendors, and its costs continue to > decline. One presumes SSD has even more "interesting" characteristics when > it comes to I/O performance and link speeds. Same with device-side > compression, which is changing things from the past. > ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
